Gemini 3.1 Pro: Pricing and Visual Coding

Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. is Google's current Pro model in Cursor. Cursor rates it medium for speed and cost and frontier for intelligence. Its distinguishing feature is native work across images and code, which Cursor calls out for UI and UX implementation from mockups, frontend development, and visual code understanding. That is a narrower and more useful reason to choose it than the generic label frontier.

Use it when the image is part of the evidence, not decoration. A screenshot showing a responsive failure, a design mockup with measurable spacing, or a visual reference paired with an existing component gives the model something concrete to compare. The model can still do ordinary repository work, but another option may be cheaper when the task has no visual input or large-context need.

Give the model the visual source, the relevant route or component, and an explicit acceptance condition. Point to the exact region in a screenshot when possible. Ask it to inspect the implementation before editing, then verify the result with the browser and a new capture. This turns the task into a visual comparison backed by code rather than a prompt asking for something to feel polished.

1. 1Attach the mockup or screenshot and identify the state, viewport, and component it represents.
2. 2Name the route and any files already known to own the UI.
3. 3State measurable constraints such as dimensions, alignment, content order, or interaction behavior.
4. 4Ask the agent to edit, run the application, and inspect the result in the browser.
5. 5Review the final screenshot, diff, responsive states, and automated checks before accepting the change.

Visual understanding does not guarantee visual correctness. A screenshot cannot show keyboard behavior, loading states, reduced motion, or a layout at every width. Include those checks when they matter and keep a person in the final review for brand judgment. Gemini can make the first pass and test what the browser exposes; it cannot infer an unstated design system policy.

Cursor lists 200k as the model's standard context window and 1M as its maximum. The higher ceiling can support a broad repository analysis, several design artifacts, or a long tool-driven session. It should not be the default input size. Large unrelated context can hide the file that matters and raises cost before the agent writes a line of code.

Start with the feature slice and expand when a search result or dependency points outward. If the task truly needs more than 200k, record why. Cursor says input and output pricing increases when input exceeds that threshold, so crossing it changes both the amount of context and the rate applied to the request.

Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. draws from the third-party Other ModelsCursor's pool of supported third-party models from providers such as Anthropic, Google and OpenAI, with separate plan access and usage from Cursor Models. Press Enter for the full definition. pool. Cursor lists standard rates of $2 per million input tokens, $0.20 per million cache reads, and $12 per million output. The cache-write cell is blank, so treat it as unpublished rather than free. Other Models access starts with Pro. Cursor StartAn India-only individual plan centered on Cursor Models, with cloud and mobile access but without Other Models, Auto, Bugbot, Automations or the Agent SDK. Press Enter for the full definition. includes only the Cursor ModelsCursor's plan and usage pool for models it labels as Cursor Models, kept separate from the third-party Other Models pool. Press Enter for the full definition. pool, so it does not include Gemini 3.1 Pro.

Cached input is one tenth of the fresh-input rate, so a stable thread that reuses the same project context can be much cheaper than repeated fresh sessions. Switching models breaks the cache. Teams and Enterprise may also pay Cursor's separate Token Rate on directly selected third-party models and eligible Auto routes, so the model table is only one part of the effective account cost.

Choose 3.1 Pro when the task needs Google's Pro reasoning or the visual coding strengths Cursor names on its model card. Choose 3.7 Flash when throughput and output cost matter more and the reasoning demand is bounded. Flash costs $0.75 input and $3.50 output against Pro's $2 and $12, with both listing cache reads at one tenth of fresh input.

Model selection should follow evidence from the result. Run the same bounded task in fresh sessions, then compare the screenshot or artifact, final diff, checks, redirects, elapsed time, and usage. A small price difference can disappear after one failed attempt, while a stronger model that produces the same result buys nothing.

Gemini 3.1 ProGoogle's frontier model in Cursor for work that combines images and code, especially UI implementation and visual code analysis. Press Enter for the full definition. can search files and folders, use web search, read and edit files, run shell commands, control the browser, generate images, ask clarifying questions, and fetch rules inside Cursor. The model reads images when the surface provides them. Tool availability still depends on the current product, approvals, sandbox, credentials, and team policy.

Give high-impact tools a clear boundary. A browser test and screenshot are useful evidence; a production deployment or external write is a separate authorized action. The model's ability to call a tool does not decide whether that action belongs in the task.

For visual tasks, store the input image and the final capture with the review. The pair makes the model's interpretation checkable later and gives a regression report a stable reference. Without both artifacts, a reviewer is left comparing the current screen against a remembered mockup or a chat description that may omit the detail that drove the edit.

A team default should also reflect source material. If most work starts from screenshots and design files, Gemini's image handling gets used. If nearly every task is plain backend code, the same strength can sit idle while the team pays a Pro-tier output rate.